EP3079106B1 - Selecting reinforcement learning actions using goals and observations - Google Patents

Selecting reinforcement learning actions using goals and observations Download PDF

Info

Publication number
EP3079106B1
EP3079106B1 EP16164072.7A EP16164072A EP3079106B1 EP 3079106 B1 EP3079106 B1 EP 3079106B1 EP 16164072 A EP16164072 A EP 16164072A EP 3079106 B1 EP3079106 B1 EP 3079106B1
Authority
EP
European Patent Office
Prior art keywords
action
neural network
goal
observation
score
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
EP16164072.7A
Other languages
German (de)
English (en)
French (fr)
Other versions
EP3079106A3 (en
EP3079106A2 (en
Inventor
Tom Schaul
Daniel George Horgan
Karol GREGOR
David Silver
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
DeepMind Technologies Ltd
Original Assignee
DeepMind Technologies Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by DeepMind Technologies Ltd filed Critical DeepMind Technologies Ltd
Publication of EP3079106A2 publication Critical patent/EP3079106A2/en
Publication of EP3079106A3 publication Critical patent/EP3079106A3/en
Application granted granted Critical
Publication of EP3079106B1 publication Critical patent/EP3079106B1/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks

Definitions

  • This specification relates to selecting actions to be performed by a reinforcement learning agent in relation to a robot interacting with the environment to accomplish a specific task.
  • Reinforcement learning agents interact with an environment by receiving an observation that characterizes the current state of the environment, and in response, performing an action from a predetermined set of actions. Some reinforcement learning agents use neural networks to select the action to be performed in response to receiving any given observation.
  • a neural network architecture for an agent which has multiple internal goals is described in Crabbe & Dyer, "Goal Directed Adaptive Behavior in Second-Order Neural Networks: The MAXSON family of architectures", Adaptive Behavior, 2000, 8(2), 149-172 .
  • Neural networks are machine learning models that employ one or more layers of nonlinear units to predict an output for a received input.
  • Some neural networks are deep neural networks that include one or more hidden layers in addition to an output layer. The output of each hidden layer is used as input to the next layer in the network, i.e., the next hidden layer or the output layer.
  • Each layer of the network generates an output from a received input in accordance with current values of a respective set of parameters.
  • a reinforcement learning system can effectively and directly consider target states that a reinforcement agent is attempting to reach in selecting actions to be performed by the agent in response to received observations when the environment is a real-world environment and the agent is a robot interacting with the environment to accomplish a specific task. Additionally, the reinforcement learning system can effectively select actions for goals that were not encountered during learning. In selecting actions, the reinforcement learning system can take advantage of a shared structure between the space of goals and the space of observations.
  • This specification generally describes a reinforcement learning system that selects actions to be performed by a reinforcement learning agent interacting with an environment.
  • the agent receives data characterizing the current state of the environment and selects an action from a predetermined set of actions to be performed in response to the received data.
  • Data characterizing a state of the environment will be referred to in this specification as an observation.
  • the agent interacts with the environment to attempt to reach a predetermined set of target states of the environment.
  • Data characterizing a state from the set of target states will be referred to in this specification as a goal.
  • the environment is a real-world environment and the agent is a robot interacting with the environment to accomplish a specific task; each target state can be a state in which a subtask has been completed by the robot, e.g., when the robot has moved an item from one location to another or has otherwise made progress toward completing the task.
  • FIG. 1 shows an example reinforcement learning system 100.
  • the reinforcement learning system 100 is an example of a system implemented as computer programs on one or more computers in one or more locations in which the systems, components, and techniques described below are implemented.
  • the reinforcement learning system 100 selects actions to be performed by a reinforcement learning agent 102 interacting with an environment 104.
  • the reinforcement learning system 100 receives observations, with each observation characterizing a state of the environment 104, and, in response to each observation, selects an action from a predetermined set of actions to be performed by the reinforcement learning agent 102.
  • the goal neural network 110 is a neural network that is configured to receive a goal and to process the goal to generate a numeric representation of the goal.
  • the numeric representation of the goal is a vector from an n-dimensional vector space of embeddings.
  • the observation neural network 120 is a neural network that is configured to receive an observation to generate a numeric representation of the observation.
  • the numeric representation of the observation is also vector from the n-dimensional vector space of embeddings.
  • the observation neural network 120, the goal neural network 110, or both are multi-layer perceptron (MLP) neural networks.
  • MLP multi-layer perceptron
  • the observation neural network 120 and the goal neural network 110 share one or more parameters.
  • the parameters of the first layer of the observation neural network 120 and the goal neural network 110 may be shared.
  • the action score neural network 130 is a neural network that is configured to receive a combined representation of a goal and an observation and to generate a respective score for each action in the set of actions.
  • the action score neural network 130 is a neural network that is configured to receive a combined representation and an action and to generate a score for the action.
  • the combined representation is generated by the reinforcement learning system 100 from the numeric representation of the goal and the numeric representation of the observation.
  • the system maps the n-dimensional goal representation and the n-dimensional observation representation to a scalar value to generate the combined representation. Generating a combined representation will be described in more detail below with reference to FIG. 3 .
  • FIG. 2 is a flow diagram of an example process 200 for selecting an action in response to a received observation.
  • the process 200 will be described as being performed by a system of one or more computers located in one or more locations.
  • a reinforcement learning system e.g., the reinforcement learning system 100 of FIG. 1 , appropriately programmed in accordance with this specification, can perform the process 200.
  • the system receives an observation characterizing the current state of the environment (step 202).
  • the system receives a goal characterizing a target state from a set of target sets of the environment (step 204).
  • the system provides an interface that allows a user to select a target state. Once selected, a goal characterizing the target state will be used by system in selecting actions to be performed by the agent until the user switches the target state through the interface.
  • the system receives the state from another system, e.g., a system that selects the target state to be reached from the set of target states, e.g., using a policy that selects the target state based on the current observation.
  • the system generates a combined representation of the observation and of the goal (step 206). Generating the combined representation is described in more detail below with reference to FIG. 3 .
  • the system generates a respective score for each action in the set of actions from the combined representation by processing the combined representation using an action score neural network (step 208).
  • the system processes the combined representation using the action score neural network to generate the respective scores.
  • the system processes the action and the combined representation using the action score neural network to generate the respective score for the action.
  • the system selects the action to be performed by the agent in response to the observation using the respective scores for the actions in the predetermined set of actions (step 210).
  • the system selects the highest-scoring action, i.e., the action having the highest score of any of the actions in the set of actions, as the action to be performed by the agent.
  • the system may at times select an action other than the highest-scoring action as the action to be performed. For example, during learning the system may select a random action from the set of actions with probability ⁇ and select the highest-scoring action with probability 1 - ⁇ .
  • FIG. 3 is a flow diagram of an example process 300 for generating a combined representation of a goal and an observation.
  • the process 300 will be described as being performed by a system of one or more computers located in one or more locations.
  • a reinforcement learning system e.g., the reinforcement learning system 100 of FIG. 1 , appropriately programmed in accordance with this specification, can perform the process 300.
  • the system processes the observation using the observation neural network to generate a numeric representation of the observation (step 302).
  • the system processes the goal using the goal neural network to generate a numeric representation of the goal (step 304).
  • the system generates a combined representation from the numeric representation of the observation and the numeric representation of the goal (step 306).
  • the system may generate the combined representation from the numeric representations of the observation and of the goal in any of a variety of ways.
  • the system may generate the combined representation by computing a dot product between the numeric representation of the observation and the numeric representation of the goal.
  • the system may generate the combined representation by processing the numeric representation of the observation and the numeric representation of the goal using a combining neural network that is configured to receive the numeric representation of the observation and the numeric representation of the goal and process the numeric representation of the observation and the numeric representation of the goal to generate the combined representation.
  • the system trains the observation neural network, the goal neural network, the action score neural network and, optionally, the combining neural network.
  • the system directly trains the neural networks jointly using an end-to-end training procedure, i.e., by obtaining labeled training data and performing a neural network training technique, e.g., stochastic gradient descent, to backpropagate errors from the action score neural network through the observation neural network and the goal neural network.
  • a neural network training technique e.g., stochastic gradient descent
  • the system trains the neural networks using a two-stage training procedure based on matrix factorization.
  • FIG. 4 is a flow diagram of an example process 400 for training an action score neural network, a goal neural network, and an observation neural network.
  • the process 400 will be described as being performed by a system of one or more computers located in one or more locations.
  • a reinforcement learning system e.g., the reinforcement learning system 100 of FIG. 1 , appropriately programmed in accordance with this specification, can perform the process 400.
  • the system trains the action score neural network to determine trained values of the parameters of the action score neural network and to determine idealized goal embeddings and idealized observation embeddings (step 402).
  • the system determines targets for the training of the action score neural network using a Horde architecture.
  • a Horde architecture a different neural network is maintained for each goal from the set of goals.
  • the neural network for each goal generates action scores based on received observations.
  • the system trains each of the neural networks in the Horde independently and off policy using conventional reinforcement learning techniques.
  • An example Horde architecture and example techniques for learning the neural networks in the Horde are described in more detail in Sutton, Richard S, Modayil, Joseph, Delp, Michael, Degris, Thomas, Pilarski, Patrick M, White, Adam, and Precup, Doina.
  • Horde A scalable real-time architecture for learning knowledge from unsupervised sensorimotor interaction. In The 10th International Conference on Autonomous Agents and Multiagent Systems-Volume 2, pp. 761-768,2011 .
  • the system can determine targets for the training of the action score neural network, i.e., action scores that should be generated by the action score neural network, based on the outputs of the learned neural networks in the Horde.
  • the system trains the action score neural network on the targets determined from the Horde using neural network training techniques, e.g., stochastic gradient descent.
  • the system determines targets for the training of the action score neural network by bootstrapping from the next states in visited transitions for each goal in a training set, e.g., using a variant of a Q-learning technique.
  • the system To determine the idealized goal embeddings and the idealized observation embeddings, the system generates a data matrix with one row corresponding to each observation encountered during the training of the action score neural network and one column corresponding to each goal encountered during the training of the action score neural network.
  • the entries in the matrix are the target action scores for the corresponding goal, observation combination.
  • the system then factorizes the matrix to determine a low-rank approximation that defines the embedding spaces for both states and goals.
  • the system trains the goal neural network on the idealized goal embeddings (step 404). That is, the system trains the goal neural network to generate numeric representations that match the idealized goal embeddings using conventional techniques.
  • the system trains the observation neural network on the idealized observation embeddings (step 406). That is, the system trains the observation neural network to generate numeric representations that match the idealized observation embeddings using conventional techniques.
  • a system of one or more computers to be configured to perform particular operations or actions means that the system has installed on it software, firmware, hardware, or a combination of them that in operation cause the system to perform the operations or actions.
  • one or more computer programs to be configured to perform particular operations or actions means that the one or more programs include instructions that, when executed by data processing apparatus, cause the apparatus to perform the operations or actions.
  • Embodiments of the subject matter and the functional operations described in this specification can be implemented in digital electronic circuitry, in tangibly-embodied computer software or firmware, in computer hardware, including the structures disclosed in this specification and their structural equivalents, or in combinations of one or more of them.
  • Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible non-transitory program carrier for execution by, or to control the operation of, data processing apparatus.
  • the program instructions can be encoded on an artificially-generated propagated signal, e.g., a machine-generated electrical, optical, or electromagnetic signal, that is generated to encode information for transmission to suitable receiver apparatus for execution by a data processing apparatus.
  • the computer storage medium can be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them.
  • data processing apparatus refers to data processing hardware and encompasses all kinds of apparatus, devices, and machines for processing data, including by way of example a programmable processor, a computer, or multiple processors or computers.
  • the apparatus can also be or further include special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit).
  • the apparatus can optionally include, in addition to hardware, code that creates an execution environment for computer programs, e.g., code that constitutes processor firmware, a protocol stack, a database management system, an operating system, or a combination of one or more of them.
  • a computer program (which may also be referred to or described as a program, software, a software application, a module, a software module, a script, or code) can be written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages, and it can be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.
  • a computer program may, but need not, correspond to a file in a file system.
  • a program can be stored in a portion of a file that holds other programs or data, e.g., one or more scripts stored in a markup language document, in a single file dedicated to the program in question, or in multiple coordinated files, e.g., files that store one or more modules, sub-programs, or portions of code.
  • a computer program can be deployed to be executed on one computer or on multiple computers that are located at one site or distributed across multiple sites and interconnected by a communication network.
  • the processes and logic flows described in this specification can be performed by one or more programmable computers executing one or more computer programs to perform functions by operating on input data and generating output.
  • the processes and logic flows can also be performed by, and apparatus can also be implemented as, special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit).
  • special purpose logic circuitry e.g., an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit).
  • Computers suitable for the execution of a computer program include, by way of example, can be based on general or special purpose microprocessors or both, or any other kind of central processing unit.
  • a central processing unit will receive instructions and data from a read-only memory or a random access memory or both.
  • the essential elements of a computer are a central processing unit for performing or executing instructions and one or more memory devices for storing instructions and data.
  • a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto-optical disks, or optical disks.
  • mass storage devices for storing data, e.g., magnetic, magneto-optical disks, or optical disks.
  • a computer need not have such devices.
  • a computer can be embedded in another device, e.g., a mobile telephone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a Global Positioning System (GPS) receiver, or a portable storage device, e.g., a universal serial bus (USB) flash drive, to name just a few.
  • PDA personal digital assistant
  • GPS Global Positioning System
  • USB universal serial bus
  • Computer-readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media and memory devices, including by way of example semiconductor memory devices, e.g., EPROM, EEPROM, and flash memory devices; magnetic disks, e.g., internal hard disks or removable disks; magneto-optical disks; and CD-ROM and DVD-ROM disks.
  • semiconductor memory devices e.g., EPROM, EEPROM, and flash memory devices
  • magnetic disks e.g., internal hard disks or removable disks
  • magneto-optical disks e.g., CD-ROM and DVD-ROM disks.
  • the processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.
  • a computer having a display device, e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor, for displaying information to the user and a keyboard and a pointing device, e.g., a mouse or a trackball, by which the user can provide input to the computer.
  • a display device e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor
  • keyboard and a pointing device e.g., a mouse or a trackball
  • Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback, e.g., visual feedback, auditory feedback, or tactile feedback; and input from the user can be received in any form, including acoustic, speech, or tactile input.
  • a computer can interact with a user by sending documents to and receiving documents from a device that is used by the user; for example, by sending web pages to a
  • Embodiments of the subject matter described in this specification can be implemented in a computing system that includes a back-end component, e.g., as a data server, or that includes a middleware component, e.g., an application server, or that includes a front-end component, e.g., a client computer having a relationship graphical user interface or a Web browser through which a user can interact with an implementation of the subject matter described in this specification, or any combination of one or more such back-end, middleware, or front-end components.
  • the components of the system can be interconnected by any form or medium of digital data communication, e.g., a communication network. Examples of communication networks include a local area network (“LAN”) and a wide area network (“WAN”), e.g., the Internet.
  • LAN local area network
  • WAN wide area network
  • the computing system can include clients and servers.
  • a client and server are generally remote from each other and typically interact through a communication network.
  • the relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.
  • Certain features that are described in this specification in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable subcombination.
  • features may be described above as acting in certain combinations and even initially claimed as such, one or more features from a claimed combination can in some cases be excised from the combination, and the claimed combination may be directed to a subcombination or variation of a subcombination.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Image Analysis (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Feedback Control In General (AREA)
EP16164072.7A 2015-04-06 2016-04-06 Selecting reinforcement learning actions using goals and observations Active EP3079106B1 (en)

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
US201562143677P 2015-04-06 2015-04-06

Publications (3)

Publication Number Publication Date
EP3079106A2 EP3079106A2 (en) 2016-10-12
EP3079106A3 EP3079106A3 (en) 2017-03-29
EP3079106B1 true EP3079106B1 (en) 2022-06-08

Family

ID=55697112

Family Applications (1)

Application Number Title Priority Date Filing Date
EP16164072.7A Active EP3079106B1 (en) 2015-04-06 2016-04-06 Selecting reinforcement learning actions using goals and observations

Country Status (4)

Country Link
US (1) US10628733B2 (zh)
EP (1) EP3079106B1 (zh)
CN (1) CN106056213B (zh)
DK (1) DK3079106T3 (zh)

Families Citing this family (48)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3696737B1 (en) * 2016-11-03 2022-08-31 Deepmind Technologies Limited Training action selection neural networks
US11580360B2 (en) 2016-11-04 2023-02-14 Google Llc Unsupervised detection of intermediate reinforcement learning goals
WO2018083672A1 (en) * 2016-11-04 2018-05-11 Deepmind Technologies Limited Environment navigation using reinforcement learning
JP6926203B2 (ja) * 2016-11-04 2021-08-25 ディープマインド テクノロジーズ リミテッド 補助タスクを伴う強化学習
EP3523761B1 (en) * 2016-11-04 2023-09-20 DeepMind Technologies Limited Recurrent environment predictors
CN117521725A (zh) * 2016-11-04 2024-02-06 渊慧科技有限公司 加强学习系统
KR20220147154A (ko) 2016-11-04 2022-11-02 딥마인드 테크놀로지스 리미티드 신경망을 이용한 장면 이해 및 생성
US11636314B2 (en) 2016-11-15 2023-04-25 Google Llc Training neural networks using a clustering loss
WO2018142212A1 (en) 2017-01-31 2018-08-09 Deepmind Technologies Limited Data-efficient reinforcement learning for continuous control tasks
WO2018142378A1 (en) * 2017-02-06 2018-08-09 Deepmind Technologies Limited Memory augmented generative temporal models
WO2018156891A1 (en) * 2017-02-24 2018-08-30 Google Llc Training policy neural networks using path consistency learning
DK3568810T3 (da) * 2017-02-24 2023-11-13 Deepmind Tech Ltd Handlingsudvælgelse til forstærkningslæring ved hjælp af neurale netværk
WO2018154100A1 (en) 2017-02-24 2018-08-30 Deepmind Technologies Limited Neural episodic control
WO2018170510A1 (en) * 2017-03-17 2018-09-20 Google Llc Mirror loss neural networks
CN110462638B (zh) 2017-03-23 2023-08-01 渊慧科技有限公司 使用后验锐化训练神经网络
EP3933713A1 (en) * 2017-04-14 2022-01-05 DeepMind Technologies Limited Distributional reinforcement learning
TWI719302B (zh) 2017-04-26 2021-02-21 美商谷歌有限責任公司 將機器學習整合至控制系統
EP3596662A1 (en) * 2017-05-19 2020-01-22 Deepmind Technologies Limited Imagination-based agent neural networks
WO2018211140A1 (en) * 2017-05-19 2018-11-22 Deepmind Technologies Limited Data efficient imitation of diverse behaviors
CN110447041B (zh) * 2017-05-20 2023-05-30 渊慧科技有限公司 噪声神经网络层
EP4273757A3 (en) * 2017-06-05 2024-02-14 DeepMind Technologies Limited Selecting actions using multi-modal inputs
WO2018224695A1 (en) * 2017-06-09 2018-12-13 Deepmind Technologies Limited Training action selection neural networks
CN110651279B (zh) * 2017-06-28 2023-11-07 渊慧科技有限公司 利用学徒来训练动作选择神经网络
JPWO2019049492A1 (ja) * 2017-09-07 2020-10-29 ソニー株式会社 情報処理装置および情報処理方法
US11625604B2 (en) 2017-10-27 2023-04-11 Deepmind Technologies Limited Reinforcement learning using distributed prioritized replay
US11604941B1 (en) * 2017-10-27 2023-03-14 Deepmind Technologies Limited Training action-selection neural networks from demonstrations using multiple losses
FI20175970A1 (en) * 2017-11-01 2019-05-02 Curious Ai Oy Setting up a control system for the target system
EP3718057A1 (en) * 2017-11-30 2020-10-07 Google LLC Neural architecture search using a performance prediction neural network
WO2019127063A1 (en) * 2017-12-27 2019-07-04 Intel Corporation Reinforcement learning for human robot interaction
JP6955105B2 (ja) * 2018-02-05 2021-10-27 ディープマインド テクノロジーズ リミテッド 方策オフ型アクタークリティック強化学習を使用する分散型の訓練
US10628686B2 (en) * 2018-03-12 2020-04-21 Waymo Llc Neural networks for object detection and characterization
CN108563112A (zh) * 2018-03-30 2018-09-21 南京邮电大学 用于仿真足球机器人控球的控制方法
EP3743853A1 (en) * 2018-05-09 2020-12-02 Deepmind Technologies Limited Performing navigation tasks using grid codes
CN109116854B (zh) * 2018-09-16 2021-03-12 南京大学 一种基于强化学习的多组机器人协作控制方法及控制系统
WO2020065024A1 (en) * 2018-09-27 2020-04-02 Deepmind Technologies Limited Stacked convolutional long short-term memory for model-free reinforcement learning
US10748039B2 (en) * 2018-09-27 2020-08-18 Deepmind Technologies Limited Reinforcement learning neural networks grounded in learned visual entities
US20210383218A1 (en) * 2018-10-29 2021-12-09 Google Llc Determining control policies by minimizing the impact of delusion
CN109598332B (zh) * 2018-11-14 2021-04-09 北京市商汤科技开发有限公司 神经网络生成方法及装置、电子设备和存储介质
US10739777B2 (en) * 2018-11-20 2020-08-11 Waymo Llc Trajectory representation in behavior prediction systems
US11755923B2 (en) 2018-11-29 2023-09-12 International Business Machines Corporation Guided plan recognition
EP3668050A1 (de) * 2018-12-12 2020-06-17 Siemens Aktiengesellschaft Anpassen einer software-anwendung, die auf einem gateway ausgeführt wird
EP3884436A1 (en) * 2019-02-19 2021-09-29 Google LLC Controlling agents using latent plans
CN110018869B (zh) 2019-02-20 2021-02-05 创新先进技术有限公司 通过强化学习向用户展示页面的方法及装置
EP3938969A1 (en) 2019-03-15 2022-01-19 3M Innovative Properties Company Determining causal models for controlling environments
CN110428057A (zh) * 2019-05-06 2019-11-08 南京大学 一种基于多智能体深度强化学习算法的智能博弈系统
CN110554604B (zh) * 2019-08-08 2021-07-09 中国地质大学（武汉） 一种多智能体同步控制方法、设备及存储设备
CN110516389B (zh) * 2019-08-29 2021-04-13 腾讯科技（深圳）有限公司 行为控制策略的学习方法、装置、设备及存储介质
US11907335B2 (en) 2020-10-16 2024-02-20 Cognitive Space System and method for facilitating autonomous target selection

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101568347B1 (ko) * 2011-04-12 2015-11-12 한국전자통신연구원 지능형 로봇 특성을 갖는 휴대형 컴퓨터 장치 및 그 동작 방법
CN102200787B (zh) * 2011-04-18 2013-04-17 重庆大学 机器人行为多层次集成学习方法及系统
CN102402712B (zh) * 2011-08-31 2014-03-05 山东大学 基于神经网络的机器人强化学习初始化方法
CN102799179B (zh) * 2012-07-06 2014-12-31 山东大学 基于单链序贯回溯q学习的移动机器人路径规划算法
US9679258B2 (en) * 2013-10-08 2017-06-13 Google Inc. Methods and apparatus for reinforcement learning

Also Published As

Publication number Publication date
US10628733B2 (en) 2020-04-21
US20160292568A1 (en) 2016-10-06
DK3079106T3 (da) 2022-08-01
EP3079106A3 (en) 2017-03-29
CN106056213B (zh) 2022-03-29
EP3079106A2 (en) 2016-10-12
CN106056213A (zh) 2016-10-26

Similar Documents

Publication Publication Date Title
EP3079106B1 (en) Selecting reinforcement learning actions using goals and observations
US11886992B2 (en) Training reinforcement learning neural networks
US11803750B2 (en) Continuous control with deep reinforcement learning
US11783182B2 (en) Asynchronous deep reinforcement learning
US11074481B2 (en) Environment navigation using reinforcement learning
US10083169B1 (en) Topic-based sequence modeling neural networks
EP3384435B1 (en) Selecting action slates using reinforcement learning
US20240078429A1 (en) Control policies for robotic agents
US11907837B1 (en) Selecting actions from large discrete action sets using reinforcement learning
WO2017083767A1 (en) Training neural networks using a prioritized experience memory
EP3207507B1 (en) Augmenting neural networks with external memory
US11354548B1 (en) Image processing with recurrent attention
US11875258B1 (en) Selecting reinforcement learning actions using a low-level controller

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

AK Designated contracting states

Kind code of ref document: A2

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

AX Request for extension of the european patent

Extension state: BA ME

PUAL Search report despatched

Free format text: ORIGINAL CODE: 0009013

AK Designated contracting states

Kind code of ref document: A3

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

AX Request for extension of the european patent

Extension state: BA ME

RIC1 Information provided on ipc code assigned before grant

Ipc: G06N 3/08 20060101ALI20170218BHEP

Ipc: G06N 99/00 20100101ALI20170218BHEP

Ipc: G06N 3/04 20060101AFI20170218BHEP

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE

17P Request for examination filed

Effective date: 20170925

RBV Designated contracting states (corrected)

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

RAP1 Party data changed (applicant data changed or rights of an application transferred)

Owner name: GOOGLE LLC

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: EXAMINATION IS IN PROGRESS

17Q First examination report despatched

Effective date: 20190523

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: EXAMINATION IS IN PROGRESS

RAP1 Party data changed (applicant data changed or rights of an application transferred)

Owner name: DEEPMIND TECHNOLOGIES LIMITED

GRAP Despatch of communication of intention to grant a patent

Free format text: ORIGINAL CODE: EPIDOSNIGR1

RIC1 Information provided on ipc code assigned before grant

Ipc: G06N 20/00 20190101ALI20211005BHEP

Ipc: G06N 3/08 20060101ALI20211005BHEP

Ipc: G06N 3/04 20060101AFI20211005BHEP

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: GRANT OF PATENT IS INTENDED

RIN1 Information on inventor provided before grant (corrected)

Inventor name: SILVER, DAVID

Inventor name: GREGOR, KAROL

Inventor name: HORGAN, DANIEL GEORGE

Inventor name: SCHAUL, TOM

INTG Intention to grant announced

Effective date: 20211111

GRAS Grant fee paid

Free format text: ORIGINAL CODE: EPIDOSNIGR3

GRAA (expected) grant

Free format text: ORIGINAL CODE: 0009210

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE PATENT HAS BEEN GRANTED

AK Designated contracting states

Kind code of ref document: B1

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

REG Reference to a national code

Ref country code: GB

Ref legal event code: FG4D

REG Reference to a national code

Ref country code: AT

Ref legal event code: REF

Ref document number: 1497430

Country of ref document: AT

Kind code of ref document: T

Effective date: 20220615

Ref country code: CH

Ref legal event code: EP

REG Reference to a national code

Ref country code: DE

Ref legal event code: R096

Ref document number: 602016072646

Country of ref document: DE

REG Reference to a national code

Ref country code: IE

Ref legal event code: FG4D

REG Reference to a national code

Ref country code: DK

Ref legal event code: T3

Effective date: 20220726

REG Reference to a national code

Ref country code: NL

Ref legal event code: FP

REG Reference to a national code

Ref country code: SE

Ref legal event code: TRGR

REG Reference to a national code

Ref country code: LT

Ref legal event code: MG9D

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: NO

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220908

Ref country code: LT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

Ref country code: HR

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

Ref country code: GR

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220909

Ref country code: FI

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

Ref country code: ES

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

Ref country code: BG

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220908

REG Reference to a national code

Ref country code: AT

Ref legal event code: MK05

Ref document number: 1497430

Country of ref document: AT

Kind code of ref document: T

Effective date: 20220608

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: RS

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

Ref country code: LV

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: SM

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

Ref country code: SK

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

Ref country code: RO

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

Ref country code: PT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20221010

Ref country code: EE

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

Ref country code: CZ

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

Ref country code: AT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: PL

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

Ref country code: IS

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20221008

REG Reference to a national code

Ref country code: DE

Ref legal event code: R097

Ref document number: 602016072646

Country of ref document: DE

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: AL

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

PLBE No opposition filed within time limit

Free format text: ORIGINAL CODE: 0009261

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: NO OPPOSITION FILED WITHIN TIME LIMIT

26N No opposition filed

Effective date: 20230310

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: SI

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

P01 Opt-out of the competence of the unified patent court (upc) registered

Effective date: 20230505

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: IE

Payment date: 20230419

Year of fee payment: 8

Ref country code: FR

Payment date: 20230421

Year of fee payment: 8

Ref country code: DK

Payment date: 20230421

Year of fee payment: 8

Ref country code: DE

Payment date: 20230427

Year of fee payment: 8

Ref country code: CH

Payment date: 20230502

Year of fee payment: 8

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: SE

Payment date: 20230421

Year of fee payment: 8

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: BE

Payment date: 20230424

Year of fee payment: 8

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: GB

Payment date: 20230418

Year of fee payment: 8

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: LU

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20230406

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: MC

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: MC

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

Ref country code: IT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20220608

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: NL

Payment date: 20240425

Year of fee payment: 9